An R package, alleleIntegrator, that integrates whole genome sequencing of tumour DNA and single cell RNA sequencing of a tumour to identify which single cells are cancer derived. As described in this paper.
The R package should be installable in the usual way with devtools
devtools::install_github('constantAmateur/alleleIntegrator')Most of alleleIntegrator won't work without also installing some external dependencies (see below).
You can satisfy the external dependencies of alleleIntegrator the easy way, or the hard way.
If you are running on a machine where singularity is installed, the main dependencies (alleleCount and bcftools) will automatically be satisfied. You shouldn't need to do anything. Although you should consider installing the optional dependencies.
If singularity is not installed and you cannot install it, you will need to make sure each of the required binaries is available. The main binaries are:
- alleleCount Needed for almost everything. Gets counts of A,C,G, and T at specified locations in BAM files.
- bcftools Needed to call hetSNPs or similar.
bcftools can be installed using apt install bcftools or brew install bcftools. Installing alleleCount is more involved, and you should follow the documentation here.
In addition to the two main external dependencies, alleleCount and bcftools, alleleIntegrator can be enhanced by installing the following:
- parallel Without this, only one core will be used at a time. Available by default on most systems, but install with
apt install parallelorbrew install parallel - ASCAT - Used only in
generateCoverageAndBAF, and is optional even then, to define copy number changes. Install withdevtools::install_github('VanLoo-lab/ascat/ASCAT')
There are four basic steps to using alleleIntegrator to identify cancer transcriptomes:
You would usually obtain this from running a copy number caller, such as ASCAT. alleleIntegrator needs a GRanges object with chromosome segments and columns matNum, patNum and tumFrac = matNum/(matNum+patNum).
Use the findHetSNPs function on the whole genome sequencing BAM. These can also be manually provided if you know them from some other source.
Use the phaseSNPsFromCN function on the whole genome sequencing of tumour BAM. You will also need to provide the heterozygous SNPs and copy number segment definitions determined in the previous steps.
Finally, the 10X BAM files are interrogated to determine which allele is expressed at each phased SNP in each copy number segment using getAllelicExpression. These counts should then be filtered using filterCells and the model fit by running calcASE, calcOverDispersion and abbSegProb.
Look at exampleRun.R for a demonstration how to use alleleIntegrator to identify cancer cell transcriptomes. This example starts from whole genome sequencing of tumour/blood from a Neuroblastoma from this paper along with cellranger mapped BAMs containing 10X 3' expression from the tumour.
This example should provide a template for how to apply this package. However, it does have some eccentricities that are specific to this sample. In particular, the DNA BAM files have been mapped against hg19, while the scRNA data is mapped against GRCh38. As such, it is necessary to perform a liftover to convert phased, heterozygous SNP coordinates between the references.
As an optional first step of this example, the matchBAMs function is used to compare the genotype of all DNA and RNA BAM files. This step is completely optional, but experience indicates that sample mixups are depressingly common and it is sensible to verify that the samples that you specify really are related in the way that you expect.
Also demonstrated in this example is how to aggregate cells into clusters and apply inference at the cluster level. Usually there is sufficient information to identify the presence/absence of the cancer genotype in each cell. But in some circumstances, it can be useful to assume that all cells in a cluster are of the same genotype and treat each cluster as if it were a single cell. If the cells in a cluster are of various genotypes, this will no improve anything, but if this assumption holds than the power to detect allelic imbalances is boosted by aggregating counts across all cells in a cluster.
If you use this package in your work, please cite: Trinh, M.K, et al., Precise identification of cancer cells from allelic imbalances in single cell transcriptomes, BiorXiv, 2021
